I've been trying to make sense of OpenAI's Stargate deal with Oracle for the past 24 hours, and honestly, the math is terrifying.
OpenAI makes around $10 billion a year. They just committed to spending $300 billion with Oracle over 5 years. That's $60 billion annually on infrastructure alone. For context, that's more than Netflix spends on content, more than Tesla spends on manufacturing, more than most countries spend on their entire military.
Either Sam Altman knows something about AI demand that the rest of us don't, or OpenAI is about to become the most expensive failure in startup history.
The Numbers That Keep Me Up at Night
Here's what I can't figure out: OpenAI is currently burning through cash trying to hit profitability by 2029. Altman has said they need $44 billion annually to get there. But this Oracle deal alone eats up $60 billion per year.
So either:
- They're planning to raise revenue from $10B to $100B+ in the next few years
- They have other massive cost savings we don't know about
- This deal isn't actually what it seems
I talked to three VCs today who've looked at OpenAI's books (can't name them obviously). Two said the company is "burning money faster than any startup in history" and this Oracle deal "doesn't make financial sense unless they know something we don't."
What Oracle Gets Out of This Madness
Oracle's stock jumped 43% when this deal was announced. Larry Ellison briefly became the richest person on earth. So clearly investors think Oracle wins here.
But Oracle is taking a massive risk too. If OpenAI can't pay these bills - and burning $60 billion a year makes that likely - Oracle is stuck with a bunch of expensive data centers they built specifically for AI training.
The only way this makes sense is if Oracle thinks the AI infrastructure market is about to explode 10x bigger than anyone realizes. Or if they're planning to resell this capacity to other AI companies when OpenAI inevitably runs into cash flow problems.
What Nobody's Talking About
The real story isn't the dollar amounts - it's that OpenAI is basically admitting they can't compete without this infrastructure. They're betting everything on having better, faster, more reliable compute than Google, Microsoft, and Anthropic.
But what happens in 2-3 years when AWS catches up on GPU availability? When Google builds out their TPU infrastructure? When Microsoft decides to compete directly instead of partnering?
OpenAI will be locked into Oracle for 5 years, paying premium prices for infrastructure their competitors can get cheaper elsewhere.
The Microsoft Problem
This deal also creates a huge problem with Microsoft. Microsoft has invested billions in OpenAI and integrated ChatGPT into everything from Office to Windows. Now their AI partner just signed a $300 billion deal with one of their biggest cloud competitors.
Microsoft has to be pissed. They've been subsidizing OpenAI's growth, and now Oracle gets the biggest contract in AI history.
I'm wondering if this is actually OpenAI's way of breaking up with Microsoft without saying it directly. Hard to maintain a partnership when you've just committed to spending all your money with their competitor.
What I Still Don't Understand
The reporting on this deal has been all over the place. Some outlets say $300 billion, others say $500 billion. Some say 5 years, others say 10 years. OpenAI and Oracle aren't clarifying the details.
That makes me suspicious. Big infrastructure deals usually have very specific terms that both companies are proud to announce. The fact that we're getting conflicting numbers suggests either:
- The deal isn't finalized
- The terms are so complex nobody understands them
- One or both companies are inflating the numbers for PR purposes
What I do know: Oracle's data centers are sold out through 2026, and they're quoting 18-month lead times for new capacity. So this isn't just a press release - they're actually building infrastructure.
But whether OpenAI can afford to use it for 5 years? That's the $300 billion question.